Inside | Real news, curated by real humans
Inside AI

Inside AI (Jul 7th, 2019)

Happy Sunday and welcome to the weekend edition of InsideAI.  If you like the newsletter, please forward it to a friend so they can subscribe.  I'm Rob May, CEO of Talla, and investor in over 60 early stage AI startups.  If you have an early stage AI company raising money, please reach out.

Here are the most popular articles from the weekday issues of InsideAI:

More companies are expected to use AI to manage their IT infrastructure. More data about networks and infrastructure, higher computing power, and more advanced algorithms have led the charge in so-called "self-driving" or "self-healing" IT. Some examples of companies using the tech include Adobe Inc., which developed an AI program using open-source technology that automates core IT tasks so employees don't have to do them, and Hitachi Vantara, a subsidiary of Hitachi Ltd., which uses AI to self-correct airflow and temperatures in data centers. - WSJ

Marketing experts are warning about AI's ability to churn out fake spam results in Google, making it difficult for humans and search engines to distinguish it from real content. The AI-generated text is cheap to generate and includes relevant keywords, but its content — such as references to people or products — doesn't actually exist. The content marketing agency Fractl is offering demonstrations of this AI-enabled text generation as it seeks to draw attention to the issue. The company used the open source tool Grover, made by the Allen Institute for Artificial Intelligence, to write fake blog posts that show up in SEO results. “Blackhats will use subversive tactics to gain a competitive advantage," Fractl partner Kristin Tynski said. - THE VERGE

The Irish startup Nuritas says it could discover four new healthcare ingredients using machine learning within the next 18 months. The company's success rate in the drug discovery process is much higher than others in the pharmaceutical industry. In collaboration with German chemical company BASF, Nuritas discovered an ingredient that helps treat inflammation, which is expected to be released in sports nutrition products by the end of this year. “We believe not only that we have launched the only healthcare ingredient found through AI, but we will in fact launch the second, third, fourth, and fifth within a 12–18 month period as well,” CEO Emmet Browne said. The company has raised $65 million so far from investors including Salesforce founder Marc Benioff and U2’s Bono and the Edge. - YAHOO FINANCE UK

  • Email gray
  • Permalink gray

Today's commentary is written by Brian Clark from Ascent.

Knowledge as a Service: A new business model

As digital services proliferate, we’re entering a new age of commercial opportunity: one where the digital world increasingly reflects changes similar to the physical world of the past 100 years.

The parables to blue collar work are striking. The building of physical widgets (i.e. a car, a cup, or a cell phone) have become increasingly efficient over the last 100 years. The difference between a Ford Factory in the 1910s and the Tesla factory of the 2010s is dispositive of this shift. The efficiency gains between the two (an exponential decrease in the marginal cost of the offered product) is now taking hold in digital services as they become productized.

As we invest fixed costs into digital machines (e.g. algorithms), a decrease in the marginal cost of digital products will follow a similar path. However, the lack of physical cost or production time requirements in digital products (algorithms produce outputs almost immediately) will mean even higher profit margins. The tradeoff in these models which have lower variable costs is higher up-front fixed costs.

We are quickly shifting from a variable cost economy to a fixed cost production economy where such investment is prioritized, resulting in higher and more permanent returns to capital. This transformation has led to multiple new types of business models, ranging from SaaS to Knowledge-as-a-Service.

Business Type

Build Costs

Distribution Costs

Acquisition Costs

Comment

Service Business

-

VC

VC

Produce Physical or Knowledge Goods

Widget Business

VC

VC

VC

Produce Physical Goods

SaaS (Software-as-a-Service)

VC

-

VC

Traditional Software Business

DaaS (Data-as-a-Service)

-

-

VC

Data (Low “T”)

IaaS (Insights-as-a-Service)

-

-

VC

Insights (Medium “T”)

KaaS (Knowledge-as-a-Service)

-

-

VC

Knowledge (High “T”)

Traditional Analog Businesses

1. Service Business

Service businesses have arguably been around longer than physical widget businesses, as prior to modern construction techniques, all work was done on a service basis. These businesses distribute their wares, which can range from physical goods to knowledge. A Physical widget business that produces an inventory of 1 is a service business. Upon standardization and mass production, a service business can become one of the other types of businesses.

A service business’ form of distribution is generally by the billable hour, though fixed cost bids for output may also be negotiated (traditionally for those that have physical outputs, though this has begun to take hold in knowledge disciplines like law, medicine, and accounting).

For those that have intangible outputs like knowledge, the human capital is the input, and the constraint is the billable hour (after all, there are only so many hours someone can work).

2. Widget Business:
The traditional build/ship/sell business. Physical widget businesses resulted from service businesses who produced standardized goods. Whether it be glass bottles, computers, or cars, each unit of production has a variable cost.

This is generally riskier than a service business, as some fixed cost in modern widget businesses is required to produce a physical good in a way that leverages economies of scale, as one-off production is not economically viable. Up until modern computing, it was one of only two business types available.

Businesses have increasingly invested fixed cost in the “build” category, reducing the variable cost. For example, a machine producing bottles may significantly reduce the variable cost until it reaches only the cost of the raw materials. Regardless, the cost of the raw materials represents the absolute minimum variable cost.

Modern Digital Businesses

3. SaaS (1990s) – A distribution business

SaaS, at its core, is a delivery vehicle: it is a container that holds some form of data, insights, or knowledge, often generated by the consumer. The true brilliance of SaaS systems has been the creation of the recurring billing model for a one-time cost to extract additional rents from consumers.

Nonetheless, SaaS is nothing more than an intelligently organized container, intended to hold and distribute some set of data, produced by the customers, to the customers. While SaaS has become a moniker for any business that can charge customers an ongoing monthly “service” fee (thus the software-as-a-service moniker), the unit margins can approach 90%+. These businesses bank on high switching costs to keep profitability high, while commoditization and the decrease of switching costs will drive margins significantly lower.

SaaS, at its core, is the productization of distribution, using the internet to repeatably deliver a software “product” that was previously sold on shelves via shrink-wrap licenses (resembling a widget business), and leveraging the positive externality of the internet to reach customers all over the world. This centralized infrastructure cost removes distribution costs. As a result, these businesses monthly have a variable cost for each product they build (many only build one), and a variable cost to acquire customers.

The marginal returns to distribution, the hallmark of the SaaS model, have drastically reduced the profitability of SaaS businesses. In addition, the rise of open-source businesses creating free products that are freely distributed over the internet has caused traditional SaaS businesses to face increasing competition (the difference in cost between a SaaS product for sale and a SaaS product for free is made up of (1) quality differences, and (2) societal goodwill premiums. As the cost of creating software decreases, and quality converges, the social premium required will shrink.

Consider an example: the production of a content-management system that cost $1,000,000 would be prohibitively expensive to offer as a free product. The production of a cloud-based content management system that costs $40,000 of labor cost may easily be taken up by an individual with the requisite skill. This competition has also created the “freemium” model, which seeks to acquire customers at a free level, and offer more valuable products on top.

4. DaaS (Data as a Service) – 2000s

As the software SaaS market matured and prices decreased in such a competitive market (where price is equal to marginal cost), firms began to seek opportunities to increase margin again.

Data become the next entry point (for example, a software system holding price reference data). These data-as-a-service models may be sold via a containerized set of data, or input into a piece of software. Nonetheless, the value is in the data produced, not the container it sits in.

This business was the hallmark of the 2000s, as global connectivity increased, and the cost of collecting and cleaning data become the prominent cost incurred by businesses shifting away from distributing solely SaaS products.

However, unlike SaaS business, the lack of a fixed cost container reduced the costs necessary to operate a business in this realm. Data can go into any container, ranging from a SaaS product, to a Microsoft Excel sheet, to an API.

By 2010, this lack of a barrier to entry led to many companies that not only could operate in the data aggregation and distribution business, but increasingly operate across verticals and industries.

The marginal returns to data have decreased as this cost structures have decreased. Little to no domain expertise is required to operate a data-as-a-service business, and as a result, much of data collection and aggregation is commoditized. As the marginal cost decreases towards zero, the revenue firms can collect trends towards zero as well. On a forward-looking basis, data-as-a- service businesses do not yield the returns necessary to invest in for the outcome levels many investors are seeking.

Putting this model into a framework will allow us to compare it to the remaining two business models more easily. A common concept with data pipelining is called “ETL”, which stands for extract, transform, and load. In DaaS business model, the amount of “transformation” of the collected data is essentially zero. Where T = 0 with some data set, you have a DaaS business.

The primary difference between a DaaS business and a SaaS business: in a SaaS business the firm doesn’t make its money primarily from content, but instead from distributing that content.

Explain transformation to digital business where Build becomes a FC instead of a VC

5. IntaaS (Insights as a Service) (2010s)

IntaaS is the business stage we are currently in. Data can be sliced, diced, analyzed, and cut into many different views, revealing profound insights. These insights are now packaged and sold to consumers.

Some firms attempt to provide insights on proprietary sets of data, thus reinforcing their “data moat,” and allowing them to collect higher rents. Their ability to provide insights on their proprietary data is the key output.

The data may be proprietary, but the value is increasingly on the insights that can be provided. These businesses are disrupted by one of two outcomes: (1) when the data isn’t proprietary, and multiple firms can enter and create the same insights, or (2) as sets of data become commoditized.

In the second example, consider a firm with a set of proprietary medical data on a random sample of 1 million people. Statistically, it is irrelevant which set of 1 million people is used, the insights would be the same when discussing population statistics (vs. sample statistics). This weakens the value of IntaaS businesses quickly.

If the size of the data set on which insights are derived is too large to be repeated from a population statistic standpoint, these businesses will still provide value and are still investable. Most today are still in this bucket, though this will decrease over time.

This business model is the first step in the productization of knowledge: firms are using raw data, applying an algorithm (though at a very low level), and extracting some insight via a transformation. When applying the ETL framework, these are businesses with a low “T” step (though it is required to be materially above zero).

KaaS (Knowledge as a Service) (2020s)

As the “T” increases, we enter the realm of knowledge businesses. In a standard low-level human knowledge process, we often apply human capital to execute a repeatable process. Much like the factories of the early 1900s, these assembly lines of knowledge have a set of repeatable knowledge steps.

In these businesses, both distributing a product (SaaS), and building a product (KaaS) are one- time fixed costs. In the latter, firms will “build a product, that builds knowledge products.” The exponential returns to this will be materially higher than SaaS businesses alone, as only variable cost of businesses like these are that of acquiring and keeping customers (see chart, above). This results in attractive unit margins, higher than that of DaaS or IntaaS businesses. In summary, these firms effectively productize knowledge.

In the ETL Framework above, these businesses have large “T” values, and the outputs of these machines mimic human-level knowledge outputs (whether it be in numerical or word-based output).

When AI is used to execute the “T” values noted above, the results increase even more significantly. Computers handle speed, and AI handles complexity. The issue with knowledge work with a large “T” isn’t economic, it’s temporal: the value exists, it simply takes too long for humans to build manually. AI (a digital machine in a cognitive factory) will automate much of the manual work, and machine outputs will mimic human outputs, but at much faster speeds.

The larger the “T” value that is automated, the larger the up-front fixed cost, and the larger the margin opportunity.

Building a KaaS Business

Building a knowledge business requires new metrics. SaaS metrics regulate product adoption and usage, but are insufficient for the proper analysis of a knowledge business. Knowledge businesses are borne from the identification of any “Repeatable Knowledge Process” (RKP). We need new measurements for progress when building one of these ventures:

1. Knowledge Market Fit

This is a form of product-market fit. The output of your knowledge machine must be akin to the human output. There is little tolerance for variance, as knowledge businesses initially will struggle for adoption if there the outputs are even marginally different. Achieving knowledge-market fit is what happens when the outputs created by your machines are sufficiently similar to the outputs a human would otherwise create or need.

Adoption with technology is largely a function of trust. Establishing trust is an important factor in any new product that utilizes cutting-edge forms of automation, from manufacturing automation to self- driving cards. The establishment of trust is a material hurdle to adoption and scaling. When done properly, it can help accelerate customer acquisition.

Beware of data creep, which is far costlier than product scope creep. As an example, expanding just one step in a knowledge process will require your humans to create an additional data point for your algorithms that may require thousands of additional analysis points. Unless you have perfect knowledge market fit, you will waste untold amounts of capital in the form of human training time for your machines for data assets that won’t be useful for your customers.

2. Unit Margin per Knowledge Unit (define, cost out, automate, monetize)

The unit margin for each knowledge widget is important analysis to understand how many customers you must acquire to begin amortizing the cost. Unit margins in the high 90s will be commonplace (at least until prices are competed down). These margins are important early on to validate the amount of up-front fixed cost investment necessary to build these businesses.

3. The “T” value

As described above, the “T” value in the ETL framework will drive the primary difference between DaaS, IntaaS, and KaaS. It will be your competitive advantage, your “intelligence moat”, and your primary differentiator. On a meta level, the “T” value represents the cognitive processes a human takes to build knowledge outputs. This is the primary reason why we are now in the era of the competitive data model.

4. Contribution Margin as a Metric

In early days, contribution margin growth will be an important long-term predictor of commercial success (since you’re investing more fixed cost into automation up front, the returns will come in the form of contribution margin first). While product-market-fit and knowledge-market fit will be early seed- stage drivers, no amount of fit and/or repeatable sales strategy will matter if contribution growth isn’t growing significantly.

Example #1 – Cable Companies

Comcast is a quintessential “SaaS” business: the company, like any cable company, spent significant funds installing infrastructure and cable lines. Most utilities operate in a similar manner, and their business exists to recoup the single product cost: the provision of cable. The content itself was licensed to third parties, with Comcast negotiating contracts with providers of content to distribute that to its customers and charging them a monthly recurring fee.

Conversely, Netflix was a Widget business that converted to a DaaS business. It’s providence as a mail- order DVD company had lower up-front fixed costs, allowing it to earn substantial revenues prior to flipping to a digital-first business. Leveraging the internet to distribute its increasingly digitalized content (meaning no Distribution Costs), it is able to massively increase its margins. As Netflix increasingly creates its own content, it is setting the stage for becoming a KaaS business. It is likely that the scripted content that Netflix is purchasing (a fixed cost) will be fed to machines which will learn and replicate these shows (the customer profiles it builds for each consumer’s preference only further corroborates this point). This manually created content will serve as training data for a set of creative algorithms that generate tv show plots.

As they start to algorithmically write shows (all of the data they’ve collected in stage one is training data for a set of algorithms), the margins will get increasingly profitable.

Example #2 – Stitch Fix

Traditional retail stores produce physical widgets: clothes. The high fixed costs of a brick & mortar widget company have been under assault for years (the Amazon effect), by DaaS companies (companies that specialize in selling products through the internet only). These DaaS companies have eliminated both fixed costs (brick & mortar), and variable costs (customers pay shipping) in order to sell their products. A digital retailer selling its clothing online is a good example of this.

Stitchfix is the evolution of this premise: they invested a fixed cost amount of capital (and likely a large amount) into a set of algorithms to create knowledge about each user. This targeting creates a better fit, and thus reduces the volume of misfits, returns, etc. Stitchfix has thus turned a variable cost (rate of return per customer) into an up-front fixed cost which builds knowledge about its customer, and thus massively reduces variable cost.

This “knowledge” is the widget they are actually selling: optimal fit.

Example #3 – Education

Education is similar retailers: traditional educational structures required localized attendance in brick & mortar establishments.

The advent of MOOCs (Massive Online Open Courseware) has been about one concept: distribution. MOOCs are “SaaS” businesses: they take external knowledge built manually, courseware and lectures, and distribute it via an online portal on a recurring basis.

As MOOCs mature, there will be opportunities to personalize educational content to support an individual, and efforts are underway to create targeted educational tracks for students. This shift will move MOOCs from SaaS businesses into the digital age, into becoming KaaS businesses. For now, they simply remain an efficient way to distribute previously created (or commissioned) content.

KaaS businesses that have product-market fit and knowledge-market fit should be judged on margin growth (and speed of margin growth) as a key growth metric.

What this means

Society is just beginning the shift towards the KaaS business. Up-front investments that yield productive first-to-market traction will mean more long-term revenue. The stickiness of KaaS businesses should be far superior to SaaS businesses when fully integrated.

The impact of this is what has been discussed above: more up-front fixed cost, more back-end margin. Knowledge businesses will also become smarter over time (knowledge is an amorphous term: automation is a function of variance and bias). As training data sets increase, and algorithms become smarter, knowledge will be easier to generate.

The application of this business allows for industry-agnostic scaling. Knowledge work automated in one industry can likely be automated in the next: automation is about domain expertise and the process of analyzing, not the industry itself. Companies that can quickly cross industries will generate outsized returns, and have growth potential that looks more like a step-function than a hockey stick.

As competition enters and pushes margins down for each of the business models above, we will approach a world where knowledge is priced competitively. Because the marginal cost of knowledge will approach zero, revenues will approach zero. The somewhat (unsurprising) conclusion is that automation will mean the marginal cost of knowledge will be reduced to near zero.

Impact

The result of this shift means that knowledge will be ubiquitous. All forms of it will be available to all people in the years to come. The potential that can be unlocked from human capital will be astounding: people will be able to create, accelerate, and design, while machines will calculate, advise, and build. For the first time in history we will be able to give people all over the world access to knowledge for free.

These knowledge machines will allow us to massively expand our abilities in areas like medicine, law, finance, and other knowledge disciplines. Capitalism as a system is intended to benefit consumers, and as the cost of knowledge drives to zero, people will be able to freely consume it. This will result in a massive consumer surplus the likes of which this world has yet to see.

A word of caution: due to the fixed cost investment nature of these knowledge sets, returns to capital will increase, and the returns to labor will continue to decrease. This transfer of wealth from variable cost sources of human labor to fixed cost sources of technology through capital investment will create challenges as the economy transitions. Similar sources of automation happened with blue-collar manufacturing work in the 50s and 60s. It is likely we will see similar automation and disruption to knowledge work in the 2020s, and more prolifically in the 2030s.

Conclusion

The productization of knowledge will drastically transform our economy, and the shift is underway. We are entering an era in which knowledge is now a commodity: a widget that can be produced and automated like the physical goods and services that came before it. The benefits of KaaS models will be significant as they take hold in the market, with the creation of long-term value and significant amounts of margin for first-to-market innovators.

--

Thanks for reading

@robmay

  • Email gray
  • Permalink gray

Subscribe to Inside AI